factors that influence listeners’ preferred bass and ... · beats by dre studio limited edition...

Dr. Sean E. Olive DIRECTOR ACOUSTIC RESEARCH

Factors that Influence Listeners’ Preferred Bass and Treble Balance in Headphones

Todd Welti SENIOR PRINCIPAL ACOUSTIC RESEARCHER

Presented at the 139th Audio Engineering Society Convention, NYC, USA October 29-Nov. 1 2015

HARMAN International. Confidential. Copyright 2015.

We’ve been busy the past 2 years researching the perception and measurement of headphones…

Audio Engineering Society Convention Paper

Presented at the 137th Convention 2014 October 10-12 Los Angeles, CA, USA

This convention paper has been reproduced from the author's advance manuscript, without editing, corrections, or consideration by the Review Board. The AES takes no responsibility for the contents. Additional papers may be obtained by sending request and remittance to Audio Engineering Society, 60 East 42nd Street, New York, New York 10165-2520, USA; also see www.aes.org. All rights reserved. Reproduction of this paper, or any portion thereof, is not permitted without direct permission from the Journal of the Audio Engineering Society.

The Correlation Between Distortion Audibility and Listener Preference in Headphones

Steve Temme1, Sean E. Olive2, Steve Tatarunis3, Todd Welti4, and Elisabeth McMullin5

1,3 Listen, Inc., Boston, MA, 02118, USA [email protected], [email protected]

2,4,5 Harman International Industries, USA [email protected], [email protected], [email protected]

ABSTRACT

It is well-known that the frequency response of loudspeakers and headphones has a dramatic impact on sound quality and listener preference, but what role does distortion have on perceived sound quality? To answer this question, five popular headphones with varying degrees of distortion were selected and equalized to the same frequency response. Trained listeners compared them subjectively using music as the test signal, and the distortion of each headphone was measured objectively using a well-known commercial audio test system. The correlation between subjective listener preference and objective distortion measurement is discussed.

1. INTRODUCTION

There has been much research published on how a loudspeaker’s linear performance, e.g. frequency, time and directional responses, affects perceived sound quality. However, there is little research published on how non-linear distortion affects perceived sound quality. In recent years, the increasing availability and affordability of high quality headphones and personal digital music players e.g. MP3 players, has brought high quality music playback to the masses. The transducer performance is critical to listener enjoyment and Dr. Olive and others have presented research on what

they believe the target frequency response of the headphone should be for optimum sound quality [1]. The attempt of this research is to determine what level and what kind of distortion is audible and how it affects the perceived sound quality.

Five different pairs of good quality over-the-ear headphones with varying levels of distortion were objectively measured and subjectively rated for their perceived sound quality. First, each headphone was equalized to the same target frequency response. Several different kinds of distortion metrics including harmonic, intermodulation, and non-coherent distortion, were measured for each headphone. A listening test was then conducted where the five headphones were rated by eight trained listeners

2

Audio Engineering Society Convention Paper

Presented at the 138th Convention 2015 May 7–10 Warsaw, Poland

Improved Measurement of Leakage Effects for Circum-aural and Supra-aural

Headphones Todd Welti

Harman International Inc. Northridge CA, 91360, USA [email protected]

ABSTRACT

Headphone leakage effects can have a profound effect on low frequency performance of headphones. A large survey, including over 2000 individual headphone measurements, was undertaken in order to compare leakage effects on human test subjects to leakage effects of the same headphones measured on a test fixture. Ten different commercially available headphones were used, each measured on eight different test subjects and a test fixture with several sets of pinnae. Modifications to the pinnae were investigated to see if the leakage effects measured on the test fixture could be made to better match the real word leakage effects measured on human test subjects.

1. INTRODUCTION

Headphone leakage effects can have a profound effect on low frequency performance of headphones. Deviations of 20 dB or more in the headphone response can easily result from varying amounts of leakage. The effect of a leak on a closed cavity is technically well understood [1]. Even so, for many headphone designs, leakage is still the largest source of variability in perceived low frequency response. Reducing this variability would be the best solution for this problem, and various strategies have been used to minimize the variation. Given that the variation cannot be entirely eliminated, a measurement method that approximates the average response on the headphones on human subjects would be advantageous. This is the focus of the current investigation.

Some previous related studies have been made comparing different pinna sets, such as in [2]. In this study different pinnae were considered, as well as different pinna hardness, but the focus was on reproducibility of measurements. There were no measurements made on human tests subjects for comparison to the artificial pinna.

The current study includes a large number of measurements, directly comparing several pinna versions to measurements on human subjects. It includes a sizeable sampling of headphones and test subjects, including over 2000 individual measurements in all.

AES

9275

This paper was peer-reviewed as a complete manuscript for presentation at this Convention. This paper is available in the AES E-Library, http://www.aes.org/e-lib All rights reserved. Reproduction of this paper, or any portion thereof, is not permitted without direct permission from the Journal of the Audio Engineering Society.

Audio Engineering Society

Convention Paper Presented at the 139th Convention

2015 October 29–November 1 New York, USA

This paper was peer-reviewed as a complete manuscript for presentation at this Convention. This paper is available in the AES E-Library, http://www.aes.org/e-lib. All rights reserved. Reproduction of this paper, or any portion thereof, is not permitted without direct permission from the Journal of the Audio Engineering Society.

Factors that Influence Listeners’ Preferred Bass and Treble Balance in Headphones

Sean E. Olive1 and Todd Welti2

Harman International, Northridge, CA, 91329, USA 1 [email protected] 2 [email protected]

ABSTRACT

A listening experiment was conducted to study factors that influence listeners’ preferred bass and treble balance in headphone sound reproduction. Using a method of adjustment a total of 249 listeners adjusted the relative treble and bass levels of a headphone that was first equalized at the eardrum reference point (DRP) to match the in-room steady-state response of a reference loudspeaker in a reference listening room. Listeners repeated the adjustment five times using three stereo music programs. The listeners included males and females from different age groups, listening experiences, and nationalities. The results provide evidence that the preferred bass and treble balances in headphones was influenced by several factors including program, and the listeners’ age, gender and prior listening experience. The younger and less experienced listeners on average preferred more bass and treble in their headphones compared to the older, more experienced listeners. Female listeners on average preferred less bass and treble than their male counterparts.

1. INTRODUCTION Recent scientific investigations into alternative headphone target curves have found that listeners prefer them when compared to the standard diffuse and free-field headphone calibrations [1]-[4]. Olive et al. showed evidence that trained listeners preferred a headphone target response that closely matched the measured in-room steady-state response of a neutral loudspeaker calibrated in a reference listening room [3]. However, the relative levels of the bass and treble sections of the headphone target response were derived empirically rather than through formal experimentation, leaving some doubt as

to whether the bass and treble levels of the headphone target response were optimized for best sound quality. To address this question, a follow up experiment was recently conducted wherein listeners directly adjusted the relative bass and treble levels of the headphone after it was equalized at the DRP to match the in-room response of a reference loudspeaker [4]. The experiment was repeated for both loudspeaker and headphone playback conditions to determine how closely the two results matched. The average preferred bass and treble levels were 4.8 dB and -4.4 dB, respectively for headphone playback, and 6.6 dB and -2.4 dB for loudspeaker reproduction. In other words, listeners on

AES

HARMAN International. Confidential. Copyright 2015. 3

Research Motivation

• Current research indicates headphones calibrated to current standards (e.g. diffuse and free-field) are less preferred to new headphone calibrations

• Olive et al. (2014) developed a new headphone target curve based on an accurate loudspeaker in a reference listening room that was preferred to current standards and several highly-regarded headphones

• We tested it using 238 listeners in four different countries from different age groups, gender, trained vs untrained listeners


Previous Research (2014)

Brand / Model Price

Harman Target Curve Based on latest AES paper October 2013 ----

Sennheiser HD800 $1500

Audeze LCD2 (rev 2) $995

Beats by Dre Studio Limited Edition $270


Mean Preference Ratings (n=238 listeners)Pr

efer

ence

Rat

ing

0

2

4

5

7

HeadphoneHarman Target Curve HP2 HP3 HP4

2.25.05.86.4

!!

!

!


Current Study Research Questions

• What are the preferred bass and treble level settings for a headphone equalized to simulate the in-room response of an accurate loudspeaker in a reference listening room?

• What is the influence of program, listeners’ age, listening experience, gender, culture on their preferred headphone spectral balance?


Method For Deriving Harman Target Curve

Step 1: Choose an accurate loudspeaker with flat on-axis, smooth off-axis response and directivity

On-axisListening WindowFirst ReflectionsPredicted In-roomSound power

Sound Power DIFirst Reflections DI

Revel Performa F208 loudspeaker



Step 2: Using standard stereo setup equalize speaker in reference listening room to produce flat in-room response at primary listening seat; response is based on spatial average of 9 mic positions

Harman Reference Listening Room


Equalizing the In-Room Loudspeaker Response to Flat

Before EQ(Left Speaker)

After EQ(Average of

Left + Right Speaker)

left channel only

average left and right channels



Step 3: At primary listening seat measure response at ear-drum using GRAS ear simulator mounted to head; take average based on 3 head positions

\

+30∘-30∘ 0∘


Method for Deriving Harman Target Curve

Step 4: Equalize headphone to match in-room response of loudspeaker measured at the ear drum. Determine what the optimal bass and treble balance should be through Method of Adjustment.

Olive and Welti Preferred Bass and Treble Balance in Headphones

AES 139th Convention, New York, USA, 2015 October 29–November 1

Page 3 of 12

Fig. 2 shows the frequency responses of the bass and treble shelving filters set at their lower and upper values for these listening experiments: - 5 dB to +15 dB for the bass filter, and -10 dB to +5 dB for the treble filter.

Figure 2 The frequency response of the bass and treble filters set to their maximum and minimum gain values. The range of possible adjustments were sufficiently large to accommodate a wide range of listeners’ tastes based on the findings in our previous investigation [4], and some initial pilot tests. The authors also verified that the maximum gain values did not overload or distort the playback chain up to and including the headphones. The frequencies for the bass and treble filters were 105 Hz and 2.5 kHz, respectively. The rationale used in selecting these frequencies was explained in the previous study (see section 2.7 of [4]) and essentially are related to cover areas where loudspeakers and rooms acoustically interact and produce the most variability in response.

2.3. Selection and Equalization of Headphones

Ten pairs of well-matched headphones (model Sennheiser HD518) were used in these experiments. The model was a dynamic, open-back headphone having a relatively smooth and extended frequency response with low distortion. These features made it a good candidate for equalization. More importantly, it proved to have a consistent fit across listeners (confirmed by measuring it on several listeners with multiple re-seats of the headphone between measurements). Being an open-back design, it provided a controlled and predictable amount of bass leakage. Together these desirable features helped to minimize variations in the signals delivered to the listeners’ ears.

The headphone was equalized at the DRP to match the flattened in-room steady-state response of a reference loudspeaker (Revel Performa F208) calibrated in the Harman reference listening room. Although the loudspeaker had a flat anechoic on-axis response it was equalized in-situ to be flat as a baseline condition.1 Details on the loudspeaker, listening room, and equalization process are available in references [4]-[5]. Fig. 3 shows the frequency response of the headphone (dotted line) measured at the DRP using the GRAS 45 CA test fixture with KB0071 pinnae (incorporating RA0045 ear simulator meeting IEC 60318) after it was calibrated to match the flat in-room steady-state response of the Revel loudspeaker (solid line). The response is the baseline from which listeners adjusted the levels of the bass and treble filters according to preference.

Figure 3 The frequency response of the Sennheiser HD518 headphone (dotted curve) equalized and measured at the DRP to match the flattened in-room steady-state response of the Revel F208 loudspeaker (black curve) in the Harman Reference Listening Room.

2.4. Playback Level

The average playback level of the headphones was set to a moderate level of 82 dB (C-weighted, slow). This

1 Equalizing the reference loudspeaker to a flat in-room target response served as a baseline condition. In doing so it, the loudspeaker no longer sounded neutral with bass and treble adjustment. In [4], we showed that the preferred in-room target response has about a 9 dB downward slope from 20 Hz to 20 kHz.


Listening Test Setup

Task: Using the two rotating knobs on the iPad adjust the bass and treble levels according to personal preference

Sennheiser HD518 equalized to “flat” Harman Target Response

Custom iPad Test Software


Harman Headphone iPad Test Software

Bass Control Treble Control


Listening Test Method

• Using a Method of Adjustment listeners adjust the bass and treble levels of the headphone in 0.25 dB increments according to personal preference

• Repeat 5x using three programs

• Randomize starting level(s) and program order

Olive and Welti Preferred Bass and Treble Balance in Headphones

AES 139th Convention, New York, USA, 2015 October 29–November 1

Page 3 of 12

Fig. 2 shows the frequency responses of the bass and treble shelving filters set at their lower and upper values for these listening experiments: - 5 dB to +15 dB for the bass filter, and -10 dB to +5 dB for the treble filter.

Figure 2 The frequency response of the bass and treble filters set to their maximum and minimum gain values. The range of possible adjustments were sufficiently large to accommodate a wide range of listeners’ tastes based on the findings in our previous investigation [4], and some initial pilot tests. The authors also verified that the maximum gain values did not overload or distort the playback chain up to and including the headphones. The frequencies for the bass and treble filters were 105 Hz and 2.5 kHz, respectively. The rationale used in selecting these frequencies was explained in the previous study (see section 2.7 of [4]) and essentially are related to cover areas where loudspeakers and rooms acoustically interact and produce the most variability in response.

2.3. Selection and Equalization of Headphones

Ten pairs of well-matched headphones (model Sennheiser HD518) were used in these experiments. The model was a dynamic, open-back headphone having a relatively smooth and extended frequency response with low distortion. These features made it a good candidate for equalization. More importantly, it proved to have a consistent fit across listeners (confirmed by measuring it on several listeners with multiple re-seats of the headphone between measurements). Being an open-back design, it provided a controlled and predictable amount of bass leakage. Together these desirable features helped to minimize variations in the signals delivered to the listeners’ ears.

The headphone was equalized at the DRP to match the flattened in-room steady-state response of a reference loudspeaker (Revel Performa F208) calibrated in the Harman reference listening room. Although the loudspeaker had a flat anechoic on-axis response it was equalized in-situ to be flat as a baseline condition.1 Details on the loudspeaker, listening room, and equalization process are available in references [4]-[5]. Fig. 3 shows the frequency response of the headphone (dotted line) measured at the DRP using the GRAS 45 CA test fixture with KB0071 pinnae (incorporating RA0045 ear simulator meeting IEC 60318) after it was calibrated to match the flat in-room steady-state response of the Revel loudspeaker (solid line). The response is the baseline from which listeners adjusted the levels of the bass and treble filters according to preference.

Figure 3 The frequency response of the Sennheiser HD518 headphone (dotted curve) equalized and measured at the DRP to match the flattened in-room steady-state response of the Revel F208 loudspeaker (black curve) in the Harman Reference Listening Room.

2.4. Playback Level

The average playback level of the headphones was set to a moderate level of 82 dB (C-weighted, slow). This

1 Equalizing the reference loudspeaker to a flat in-room target response served as a baseline condition. In doing so it, the loudspeaker no longer sounded neutral with bass and treble adjustment. In [4], we showed that the preferred in-room target response has about a 9 dB downward slope from 20 Hz to 20 kHz.

+15 dB

-5 dB

+5 dB

-10 dB


Program Material

Program / Artist / CD / Track / Label / Release Date Details

ES / Estelle w. Kayne West / Shine/ American Boy / Atlantic Records, 2008, B00142Q7H8 / April 29 2008

Hip Hop with male and female vocal

SD / Steely Dan / Two Against Nature / Cousin Dupree / Giant Records /B00004GOXS/ February 29 2000

Jazz Pop with Male Vocal

JW / Jennifer Warnes/ Blue Raincoat /Bird on a Wire Rock Pop with Female Vocal


Long-term Spectrum of Programs


Listeners

Group Country Sample Size (n)

Gender Age of Listeners in Years Listening Experience % of Group

% of Males % of Females Median SD None A Little Medium A Lot

Harman NR USA 9 78% 22% 39.4 10.4 0 11 22 67

Harman FH USA 24 83% 17% 36.0 12.9 29 13 17 42

Harman KB Germany 72 93% 7% 38.0 9.1 33 26 25 15

Harman SZ China 20 80% 20% 31.5 6.0 10 35 5 50

Citrus College

USA 24 75% 25% 23.0 6.0 21 25 38 17

Loyola M. University

(LMU)

USA 15 93% 7% 21.0 6.0 0 20 80 0

Harris College

Canada 48 90% 10% 23.0 6.0 8 8 31 52

JBL Dealers Varied 37 100% 0% 40.0 9.3 11 14 51 24

TOTAL 249 89% 11% 32 11.4 19% 19% 32% 30%

HARMAN International. Confidential. Copyright 2015.

RESULTS


Explanation of Listening Test Graphs

Preferred Bass and Treble Levels

Rel

ativ

e L

evel

(dB

)

-2

0

2

4

6

8

Bass Treble

-1.41

6.44

Bass and Treble Filters Gain

Rel

ativ

e Le

vel (

dB)

-2

0

2

4

6

8

Frequency (Hz)

10 100 1000 10000

=

6.44 dB

-1.41 dB


Preferred Bass and Treble Levels

Rel

ativ

e L

evel

(dB

)

-2

0

2

4

6

8

Bass Treble

-1.41

6.44

Averaged across all programs and all listeners (n=249)


Effect of Program on Preferred Bass and Treble Levels

Rel

ativ

e Le

vel (

dB)

-2

0

2

4

6

8

ProgramES JW SD

-1.53-1.22-1.48

7.075.806.46

Bass Treble


Effect of Gender on Preferred Bass and Treble Levels

Rel

ativ

e Le

vel (

dB)

-4

-2

0

2

4

6

8

Bass Treble

-3.1

5.6

-1.2

6.5

Male Listeners (n=222) Female Listeners (n=27)


Effect of Listening Experience on Preferred Bass and Treble Levels

Rel

ativ

e Le

vel (

dB)

-2

0

2

4

6

8

Listening Experience

None Little Medium Amount A Lot

-1.71-1.37-1.32-1.11

5.886.406.867.00

Bass Treble


Effect of Age of Listener

Rel

ativ

e Le

vel (

dB)

-4

-2

0

2

4

6

8

Listener’s Age (years)15-25 26-35 36-45 46-55 56+

-0.6-1.7-1.7-1.6-1.1

5.05.75.86.57.3

Bass Treble


Listeners’ Country of Residence

Germany31%

China9% USA

40%

Canada21%


Country of Listener

Rel

ativ

e Le

vel (

dB)

-4

-2

0

2

4

6

8

Country of Listener

Canada China USA Germany

-1.9-1.3-1.6-1.0

5.96.66.66.7

Bass Treble


Listening Experience vs Country

Canada

52%31%

8%8%

No Experience A little Medium A Lot of Experience

USA

27%

40%

18%

14%

China

50%

5%

35%

10%

Germany

15%

25%

26%

33%


Listening GroupsRe

lativ

e Le

vel (

dB)

-4

-2

0

2

4

6

8

Listening Group

Harman

NR

Harman

FH

Harman

KB

Harman

SZ

Citrus C

ollege

LMU

Harris I

nsitu

te

JBL D

ealer

s

-1.1-1.0-2.3

0.6-1.6-1.9-2.0

-3.4

6.86.77.17.56.65.95.56.0

Bass Treble

Heavy Bass & Treble LoversModerate Bass & Treble Lovers


Comparison With Previous Studies

Rel

ativ

e Le

vel (

dB)

-5.25

-3.5

-1.75

0

1.75

3.5

5.25

7

Olive & Welti (2015) Olive et al (2014) Olive et al (2014)

-2.4-4.4

-1.41

6.64.8

6.6

Bass Treble

Playback: Headphone Headphone Loudspeaker


Preferred Headphone Response as Measured at Ear Drum Reference Point

Olive & Welti Headphone Target Curve (this study)

Olive et al. Preferred Loudspeaker and Headphone Target

AES 135th Convention, New York, NY, USA, 2013 October 17–20

Page 13 of 16

preferred headphone target curve, the DF calibration would make the headphone sound too thin and bright due to the lower bass and higher treble levels. This was reported in a previous study [6], and has been confirmed again in the current study.

Figure 17 The preferred headphone target response measured at DRP (black) based on this study. Also shown is the measured response of the loudspeaker equalized to a flat in-room target response.

4.3 Measured Responses of Loudspeaker Equalized to the Preferred Target

Figs. 18 shows the measured frequency response of the Revel F208 loudspeaker equalized to the preferred in-room target response (solid curve) found in this study.

Figure 18 The measured in-room response of the Revel F208 (solid line) equalized to the preferred target response curve. Also shown is the measured response of the loudspeaker equalized to a flat target response (dotted).

For the sake of comparison, we also show the measured response of the loudspeaker equalized to a flat in-room target response. The flat in-room loudspeaker response curve would have too much treble and not enough bass to produce satisfying results for listener as confirmed in this study and a previous one [5]. Finally, Fig. 19 shows the measured in-room response of the Revel F208 (black) equalized to the preferred target response curve. This measured response is very similar to the in-room loudspeaker target (dotted red) that was preferred by listeners in two previous studies where they evaluated different loudspeaker-room correction products [5], and different headphone target response curves [6]. Also shown in Fig. 19 is the predicted in-room response of the loudspeaker based on anechoic measurements (see Fig. 1). Above 200 Hz there is good agreement between the predicted in-room response of the loudspeaker (based on anechoic measurements with no room equalization), and the measured in-room response of the loudspeaker equalized to the preferred target response. What this tells us is that a well-designed loudspeaker shouldn’t require much equalization above the transition frequency where the room no longer dominates the quality of sounds heard. However, below 100-300 Hz, the loudspeaker will likely need equalization to deal with room mode and boundary effects, and possibly some bass enhancement to satisfy the tastes of individual listeners, and accommodate variations in the quality of program material.

Figure 19 The measured in-room response of the Revel F208 loudspeaker equalized to the preferred in-room target curve (black), the predicted in-room response of the loudspeaker (cyan) based on anechoic measurements (see Fig. 1), and the modified in-room loudspeaker target curve, RR1 (red dotted) from [6].

Olive et al. Headphone Target Curve from [4]Loudspeaker equalized to flat in-room response [4]

[4] S.E. Olive, T. Welti and E. McMullin, “Listener Preferences for In-Room Loudspeaker and Headphone Target Responses,” presented at the 135th Convention, Audio Eng., Soc., preprint 8994, (2013 October).


Conclusions (1)

• Preferred levels averaged across all programs and listeners were: Bass = 6.6 dB and Treble = -1.4 dB

• Music Program a significant factor • Listener Age is significant factor: younger (15-25 years)

listeners tend to prefer high levels of bass (+1.6 dB) and treble (+0.6 dB) than older listeners (hearing loss could be a factor)

• Listening Experience a significant factor: Less experienced listeners prefer more bass and treble


Conclusions (2)

• Females preferred 1 dB less bass and 2 dB treble than male counterparts

• Germans on average preferred slightly less bass (< 1 dB) than Chinese, Canadian, USA listeners (could be explained by higher mean age of German sample)


Some Limitations of Research

• Method of Adjustments in bass and treble not loudness compensated: some listeners may have simply boosted bass and treble because it made the music louder

• Listener sample was not randomly selected or balanced • Listener training was self-reported (exception was Harman trained

listeners) • Programs limited to 3 samples • Audiometric hearing performance was not tested (except Harman

trained listeners) • In practice, preferred bass and treble level may vary with

background noise (masking) and headphone fit (leakage)


Take-aways

• The ideal headphone target curve is a moving target due to variations in:

• Program material (“circle of confusion” / lack of meaningful standards in recording industry monitoring chain)

• Individual taste (age and experience seem to matter) • Masking effects of background noise (bass/mids are

masked) • Variation in headphone fit (bass leakage)

• We need a simple bass / treble control on headphones to compensate for these variations

THANK YOU !

For more questions contact: [email protected] [email protected]

mailto:[email protected]

mailto:[email protected]?subject=

HeadphoneTechnology

2016 AES International Conference on

August 24th - 26th, Aalborg, Denmark

aes.org/conferences/2016/headphones/Audio Engineering Society

Audio Engineering Society

HeadphoneTechnology

2016 AES International Conference on

August 24th - 26th, Aalborg, Denmark

Proposed Topics:Design

Headphone Transducers: Technology, Measurement, Microdrivers, Smart - Control, Simulation Personalization Additional Sensors Wearing and Listening Comfort

Applications

Augmented Reality Mobile Spatial Audio Personal and Assistive Listening

Binaural Techniques Up-Mixing Noise Control Monitoring and Analytic Listening

Evaluation

Headphone Quality Quality Standards Automatic Quality Evaluation Perceptual Audio Evaluation Perceptual Targets

aes.org/conferences/2016/headphones/

factors that influence listeners’ preferred bass and ... · beats by dre studio limited edition...

Documents